Professional Documents
Culture Documents
Iban Plaited Mat Motif Classification With Adaptive Smoothing
Iban Plaited Mat Motif Classification With Adaptive Smoothing
Corresponding Author:
Silvia Joseph
Faculty of Computer Science and Information Technology, Universiti Malaysia Sarawak
94300 Kota Samarahan, Sarawak, Malaysia
Email: silvia.joe8@gmail.com
1. INTRODUCTION
Borneo is widely known as a home to the world’s richest and aesthetically appealing plait work
traditions. These traditions derive from the island’s diverse cultural history and a product of creativity as well
as ingenuity of its different communities [1]. Decorative mat plaiting is a craft skill known in most
indigenous communities in Borneo. Plaited mats are used for basic floorcovering, for sitting, for sleeping and
for ritual purposes during religious ceremonies. Sleeping and ritual mats tend to be the most elaborately
decorated and may feature sacred motifs that are believed to be carrying spiritually powerful patterns that
establish link between humans and the gods [2]. Modern practice of naming plaited mat motifs has suffered
from a relative neglect in material culture studies, often resulting to a devoid of meaning. The relation
between a motif and the name by which it is called is often not relevant. Thus, it is specious to interpret
motifs only through their names [3]. Our previous work [4] investigated the use of an invariant image-based
feature descriptor for recognizing Iban plaited mat motifs. We argue that this is the best route (i.e., image-
based classification) to automate the mat motif recognition task. With modern smartphones being equipped
with camera(s) as standard, we can integrate feature into educational and edutainment app. Such apps will be
an excellent way to promote and educate the public about this invaluable cultural heritage to the masses.
This paper focuses on plaited mat work by the Iban communities. Plaited mats contain unique motifs
that are either simple or complex. The motifs are mostly stylized beyond recognition. Most motifs are plaited
based on loose categories that include natural elements such as plants, animals, firmaments, and faerie [5].
The main motif plays a dominant role for determining the specific meaning, whereas the smaller motifs only
serve as fillers for filling in the empty areas surrounding the main motif. These motifs are usually arranged
repeatedly to illustrate the mat pattern as a whole see Figure 1. Each motif has its own distinctive features.
Thus, to recognize a plaited mat motif, we need to extract discriminative as well as robust visual features.
Nevertheless, due to the diversity of appearances, variety of positions, scales and rotations of the major and
minor motifs, the recognition task is non-trivial. Furthermore, the image captured by the camera may undergo
common image deformations hence making the task harder.
Figure 1. An example of an Iban plaited mat motif, i.e., Buah Tungku. The (1) and (2); small ornament that
act as a filler, Buah Mata Punai. The image was sourced from [4]
To date, a significant amount of research had been done on the robust detection, description and
matching of invariant features related to motif and pattern classification. Features extraction algorithm and
classification methods were applied to batik motif and batik making using convolutional neural network
(CNN) model architectures [6]–[10], using multiwindow and multiscale extended center symmetric local
binary patterns (MU2ECS-LBP) [11] and [12], using scale invariant feature transform (SIFT) [13], [14] using
gray level co-occurrence matrices (GLCM) [15], [16], fine arts [17]–[19], for license plate recognition [20],
[21], and tattoo recognition [22]–[24]. Even though the reported performance was quite high but these
methods still suffer from false positives due to similar features that contain more than one pattern and noisy
background. In order to deal with the mentioned problems, we introduce a classification framework with
adaptive smoothing to remove small image features whilst protecting the discriminative structural
information (i.e., edges). The appearance of the retained edges is enhanced using morphological dilation. The
rest of this paper is organized: section 2 details out the data collection and pre-processing phase, section 3
outlines the proposed method. The results and analysis are presented in section 4 whilst the conclusion is
provided in section 5.
Iban plaited mat motif classification with adaptive smoothing (Silvia Joseph)
842 ISSN: 2252-8938
Figure 2. Examples of synthetic images, i.e., Buah Lang Antu, Buah Burung Garuda, Buah Bandau, Buah
Lintan Tanah and Buah Nabau Besundang (from left to right)
3. PROPOSED METHOD
The proposed method utilizes adaptive smoothing and morphological dilation to not only remove
noises but also to recover lost image information and to enhance image details. We start by convolving the
800×533 greyscale image with a Gaussian kernel of size 5×5, with the 1st derivation of 2D Gaussian
smoothing filter function. The purpose of this smoothing function is to destroy the small image features that
we considered as noise. The smoothed image is then converted into an edge image using Canny filter. In our
proposed classification framework with adaptive smoothing, we gradually reduce the number of remaining
edges by iteratively applying Canny filter of increasing low threshold value to the original edge image. We
use a fixed increment of 5 units. By setting a higher value for the Canny filter’s low threshold parameter, the
number of kept edges is lower as the algorithm becomes more stringent in rejecting weak candidates. We
stop the iteration once the percentage of remaining edges (PoRE) is lower or equal to the set threshold values
(i.e., 100.0%, 90.0%, 85.0%, 80.0%, 75.0%, 70.0%, 50.0%, and 20.0%). These threshold values are selected
randomly as it is possible to tune the threshold and locate the optimal value directly. These thresholds are
evaluated using a validating set. PoRE is calculated using,
Current number of detected edges using PHT
PoRE = x 100 (1)
Initial number of detected edges using PHT
We calculate the current percentage at each iteration using probabilistic hough transform (PHT) to
extract the image’s structural information. The iteration will stop once PoRE is equal or less than the target
value, as shown in Figure 3 for a flowchart of the adaptive smoothing function. This function is implemented
inside our classification framework.
To validate our classification framework, we conducted an experiment for motif image classification
against the following common image deformations, i.e., image blur, illumination changes, JPEG
compression, viewpoint changes, scale+rotation changes, and scale changes. Using our dataset, we prepared a
test sequence (i.e., query images) for each motif class. A test sequence contains five images. Each image
exhibits a gradual increment of geometric and/or photometric transformation of the initial image. Following
the recommendation by [25], we used the same 6 image deformation types as in our previous work [4]. For
each motif class, both test image (i.e., synthetic image) and the test sequence’s images were converted into
edge images at different PoRE values. The final edge images were used during the classification exercise. As
a comparison set, we also produced a second set by applying PHT on the final edge images.
In total, we have 20 motif classes x 6 image deformation types x 5 deformation levels x 8 sets of
PoRE values range of (100%, 90%, 85%, 80%, 75%, 70%, 50%, and 20%), yielding a final total of 9,600
query images. Figure 4(a) shows examples of query image produced at different PoRE values, whilst Figure
4(b) shows the corresponding query images with dilation (to improve appearance of edges). Figure 5 shows
selected query images with incremental degrees of image deformation. For the feature matching and
classification of the prepared target and query images, binary robust invariant scalable keypoints (BRISK)
[26] were used. BRISK detects corners using adaptive and generic accelerated segment test (AGAST)
algorithm [27] and perform filtering with features from accelerated segment test (FAST) corner score while
searching for maxima inside the scale space pyramid. Outlier-rejection from matched features was performed
using random sample consensus (RANSAC). RANSAC estimates the homography matrix and use that
information to eliminate any mismatches/possible false key points. We used the ready implementation of
BRISK and RANSAC in VLFeat Version 0.9.21 library [28]. This work was implemented on a Windows
laptop with the following specifications: Intel Core i7-8550U @ 1.80GHz processor and 8 GB RAM.
(a) (b)
Figure 4. From left to right, query images produced using PoRE values of 100.0%, 75.0%, and 20.0% for the
following motif classes (top to bottom row) Buah Lang Antu, Buah Burung Garuda, Buah Bandau, Buah
Tungku Lebur and Buah Nabau Besundang. Two versions are shown, i.e., (a) without dilation and
(b) with dilation
Iban plaited mat motif classification with adaptive smoothing (Silvia Joseph)
844 ISSN: 2252-8938
Fronto-parallel
5% 5% 5% surface to the right 95% 15º
10º
Fronto-parallel
35% 30% 20% surface to the left 65% 50º
10º
Figure 5. From top to bottom, selected query images with incremental degrees of image deformation. From
left to right, a column showing query images of Buah Lang Antu sequence for blur changes, Buah Burung
Garuda sequence for JPEG compression changes, Buah Bandau sequence for illumination changes, Buah
Tungku Lebur sequence sequence for viewpoint changes, Buah Nabau Besundang for scale changes, and
Buah Tangkai Gajai Besembah sequence for scale + rotation changes
Figure 6. Three motifs with high misclassification rate (a) Buah Bulan Bekarung,
(b) Buah Tangkai Gajai Besembah, and (c) Buah Mata Punai
Table 1. Number of detected BRISK+RANSAC keypoints using the optimal PoRE value of 75.0% for each
image deformation category, with and without morphological dilation
Image deformation w/o dilation w/ dilation
1st level 5th level 1st level 5th level
Blur 155 0 7882 164
Illumination 117275 50991 140871 64996
Viewpoint 457 36 2994 231
JPEG compression 82992 1300 128084 10379
Scale+rotation 397 30 1307 122
Scale 795 138 2516 527
Table 2. Comparative failure point of using BRISK+RANSAC with optimal PoRE value of 75.0%,
for each image deformation category. We show the failure point (a) w/o morphological dilation, and
(b) w/ morphological dilation
(a)
Level of image deformation Blur Illumination JPEG _Compression Viewpoint Scale+rotation Scale
Image1 FN TP TP TP FN FN
Image2 FN TP TP TP FN FN
Image3 FN TP TP TP FN FN
Image4 FN TP TP FN FN FN
Image5 FN TP TP FN FN FN
(b)
Level of image deformation Blur Illumination JPEG _Compression Viewpoint Scale+rotation Scale
Image1 TP TP TP TP TP TP
Image2 TP TP TP TP FN TP
Image3 FN TP TP TP TP TP
Image4 FN TP TP TP FN TP
Image5 FN TP TP TP FN FN
The results of keypoint matching from both query and template image were summarized in the form
of confusion matrix as shown in Figures 7(a) to 7(f). We have 20 actual classes and 20+1 predicted classes.
The additional predicted class, i.e., class unresolved, represent cases where our classification returns multiple
top matches or a similarity score is lower than the threshold. We consider such cases as a FN to penalise the
configuration. Based on the confusion matrix obtained from the classification results using our proposed
framework with optimal PoRE value of 75.0% + dilation, for Blur set as in Figure 7(a) shows the highest
number of unresolved classes, followed by scale set as in Figure 7(d), and scale+rotation set, in Figure 7(f).
Most of the results in class unresolved for blur set are due to the false top matches produced by simple motif
Buah Mata Punai and complex motifs of Buah Bulan Bekarung, Buah Kandung Nibung Berayah, Buah
Sebayan and Buah Tangkai Gajai Besembah. From our experiment, we observed that blurring is the most
destructive image deformation type (compared to the rest) as its application produces many false top
matches. There is no class unresolved for JPEG compression set, illumination set, and viewpoint set, see
Iban plaited mat motif classification with adaptive smoothing (Silvia Joseph)
846 ISSN: 2252-8938
Figures 7(b) to 7(e). Figure 8(a) shows a synthetic image of Buah Tungku Lebur, and Figure 8(b) shows the
resulting Canny edge image. At PoRE value of 75.0%, the visualized detected BRISK features after blurring
at increasing levels are shown in Figures 8(c) to 8(f). Side-by-side comparisons of visualized matching
results for selected motifs using BRISK+ RANSAC at deformation level 1 (left column) vs. level 5 (right
column) are shown in Figure 9. We include a sample visualized matching result each for the following image
deformation types, i.e., i) blur, see Figure 9(a); ii) viewpoint, see Figure 9(b); iii) scale, see Figure 9(c); and
iv) scale+rotation, see Figure 9(d). Clearly demonstrated that with our proposed method we managed to
match enough key points, even at the highest deformation level. Based on the confusion matrix, accuracy is
calculated as,
𝑇𝑃+𝑇𝑁
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 = (2)
𝑃
Where; TP= true positive, TN= true negative (by default is 0), P= population.
BBBepayung
BBBepayung
BBBekarung
BBBekarung
unresolved
unresolved
BKNB
BTGB
BKNB
BTGB
BLM
BLM
BMP
BMP
BLA
BBG
BNN
BNB
BNN
BSA
BLT
BTL
BLA
BBG
BNB
BTL
BM
BM
BA
BK
BB
BA
BK
BT
BP
BS
BB
BT
BP
BS
motif motif
BLA 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BLA 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BB 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BB 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BA 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BA 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BBBekarung 0.00 0.00 0.00 0.60 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.40 BBBekarung 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BBBepayung 0.00 0.00 0.00 0.00 0.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20 BBBepayung 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BBG 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BBG 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Actual classes
Actual classes
BKNB 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20 BKNB 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.60 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.40 BP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BK 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20 BK 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BLM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20 BLM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BLT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20 BLT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BMP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.40 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.60 BMP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.60 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.40 BM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BNB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BNB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BNN 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 BNN 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00
BS 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.60 0.00 0.00 0.00 0.00 0.40 BS 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00
BSA 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 BSA 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00
BTGB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 BTGB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00
BTL 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 BTL 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00
BT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.20 BT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00
(a) (b)
Predicted classes Predicted classes
BBBepayung
BBBepayung
BBBekarung
BBBekarung
unresolved
unresolved
BKNB
BKNB
BTGB
BTGB
BLM
BMP
BLM
BLA
BBG
BNN
BNB
BMP
BLA
BBG
BNN
BNB
BSA
BLT
BTL
Plaited mat
BSA
BLT
BTL
Plaited mat
BM
BM
BA
BK
BB
BA
BK
BB
BT
BP
BS
BT
BP
BS
motif motif
BLA 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BLA 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BB 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BB 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BA 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BA 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BBBekarung 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BBBekarung 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BBBepayung 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BBBepayung 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BBG 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BBG 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Actual classes
Actual classes
BKNB 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BKNB 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BK 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BK 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BLM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BLM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BLT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BLT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BMP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BMP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BNB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BNB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BNN 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 BNN 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00
BS 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 BS 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00
BSA 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 BSA 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00
BTGB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 BTGB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 0.00 0.20
BTL 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 BTL 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00
BT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 BT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00
(c) (d)
Predicted classes Predicted classes
BBBepayung
BBBepayung
BBBekarung
BBBekarung
unresolved
unresolved
BKNB
BKNB
BTGB
BTGB
BLM
BLM
BMP
BMP
BLA
BBG
BNN
BNB
BLA
BBG
BNN
BNB
BSA
BLT
BTL
BSA
BLT
BTL
BA
BK
BB
BA
BK
BB
BT
BP
BS
BT
BP
BS
motif motif
BLA 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BLA 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BB 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BB 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BA 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BA 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BBBekarung 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BBBekarung 0.00 0.00 0.00 0.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20
BBBepayung 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BBBepayung 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BBG 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BBG 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Actual classes
Actual classes
BKNB 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BKNB 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20
BP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BK 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BK 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BLM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BLM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BLT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BLT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BMP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BMP 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.60 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.40
BM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BM 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.20
BNB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 BNB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
BNN 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 BNN 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00
BS 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00 BS 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00
BSA 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 BSA 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00
BTGB 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 0.00 BTGB 0.00 0.00 0.00 0.20 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.60 0.00 0.00 0.20
BTL 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 0.00 BTL 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.80 0.00 0.20
BT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00 BT 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.00
(e) (f)
Figure 7. Confusion matrix obtained from the classification results using our proposed framework with
optimal PoRE value of 75.0%+dilation, for (a) blur set, (b) JPEG compression set, (c) illumination set, (d)
scale set, (e) viewpoint set, (f) scale+rotation set
Figure 8. Synthetic image of (a) Buah Tungku Lebur, (b) the resulting edge image produced by Canny filter,
detected BRISK key points when using optimal PoRE value of 75.0% at (c) blur level 1, (d) level 2, and
(e) level 3, and (f) level 5
In Figure 10, we presented the average matching accuracy achieved by the following schemes:
scheme 1: synthetic image and BRISK+RANSAC; scheme 2: synthetic image and BRISK+RANSAC with
Canny filter; scheme 3: synthetic image and BRISK+RANSAC with Canny filter and PoRE (without
dilation); and scheme 4: synthetic image and BRISK+RANSAC with Canny filter, PoRE (with dilation) the
average accuracy is 85.0%, 95.0%, 79.0% and 96.0% respectively. As for dilation vs no dilation, we
observed that with dilation enabled, the average accuracy increases by 17.0%. The optimal scheme, i.e.,
scheme 4 offers an average increase of 11.0% compared to the baseline, i.e., scheme 1, and an average
increase of 1.0% compared to scheme 2. Figure 11, illustrate the average matching accuracy using
BRISK+RANSAC at different PoRE values for each image deformation type. The optimal PoRE value is
observed to be 75.0% with a comparably higher average matching accuracy of 83.0%, 100.0%, 100.0%,
100.0%, 92.0% and 99.0% across all image deformations (blur, illumination, JPEG compression, viewpoint,
scale+rotation and scale) compared to the rest.
(a)
(b)
(c)
(d)
Figure 9. Side-by-side comparisons of visualized matching results of Buah Lang Antu, Buah Tungku Lebur,
Buah Nabau Besundang and Buah Tangkai Gajai Besembah using BRISK+ RANSAC at deformation level 1
(left column) vs. level 5 (right column). We show a sample matching result (optimal PoRE=75.0%, dilation
enabled) each for the following image deformation types, i.e., (a) blur, (b) viewpoint, (c) scale, and
(d) scale+rotation
Iban plaited mat motif classification with adaptive smoothing (Silvia Joseph)
848 ISSN: 2252-8938
Figure 10. Performance comparison of average matching accuracy using BRISK+RANSAC at different
schemes, following six image deformations
Figure 11. Average matching accuracy of BRISK+RANSAC at different percentage of PoRE+dilation values
following six image deformations
5. CONCLUSION
In this work, we presented an adaptive smoothing scheme consisting of Canny edge detector and
PHT to gradually remove smaller image structures from a given mat motif greyscale image. Dilation is then
used to restore broken edges after the application of the smoothing scheme to improve detection. Motifs are
matched using BRISK features and RANSAC against six common image deformations: combination of scale
and rotation, viewpoint, image blur, JPEG compression, scale and illumination changes. Varying degrees of
each deformation type were applied to all cropped and perspective-corrected mat motif images. The optimal
PoRE value is 75.0% with a reported 100.0% matching accuracy for JPEG compression, illumination and
viewpoint change set. Our optimal scheme achieved an average increase in accuracy of 11.0% compared to
classification using only BRISK+RANSAC on synthetic RGB images (i.e., baseline method). It is observed
that the proposed scheme improves the classification accuracy of mat motif images with complex motif
across common image deformation types.
ACKNOWLEDGEMENT
We wish to acknowledge Universiti Malaysia Sarawak (UNIMAS) for providing the necessary fund
towards the publication of this research work.
REFERENCES
[1] B. S. B, “Plaited arts from the Borneo rainforest,” Lontar Foundation, 2012, doi: 10.4000/moussons.2950.
[2] B. Sellato and J. Fogel, “Decorated mats of the peoples of the Borneo hinterland, & Tapis décorés des peuples de l’arrière-pays de
Bornéo,” Tribal Art Magazine, vol. XVII, no. 4, pp. 126–137, 2013.
[3] B. Sellato, “Variation as norm: names, meanings, and referents in Borneo basketry decoration.,” Borneo Research Bulletin, vol.
48, pp. 185–200, 2017.
[4] S. Joseph, I. Hipiny, H. Ujir, S. F. S. Juan, and J. L. Minoi, “Performance evaluation of SIFT against common image
deformations on iban plaited mat motif images,” Indonesian Journal of Electrical Engineering and Computer Science, vol. 23,
no. 3, pp. 1470–1477, 2021, doi: 10.11591/ijeecs.v23.i3.pp1470-1477.
[5] B. Sellato, “Art and identity in the plaited arts of Borneo: an introduction,” Plaited Art from the Borneo Rainforest, pp. 2–38,
2012.
[6] D. G. T. Meranggi, N. Yudistira, and Y. A. Sari, “Batik classification using convolutional neural network with data
improvements,” International Journal on Informatics Visualization, vol. 6, no. 1, pp. 6–11, 2022, doi: 10.30630/joiv.6.1.716.
[7] J. Tristanto, J. Hendryli, and D. Erny Herwindiati, “Classification of Batik motifs using convolutional neural networks,” SSRN
Electronic Journal, 2018, doi: 10.2139/ssrn.3258935.
[8] H. Prasetyo and B. A. Putra Akardihas, “Batik image retrieval using convolutional neural network,” Telkomnika
(Telecommunication Computing Electronics and Control), vol. 17, no. 6, pp. 3010–3018, 2019, doi:
10.12928/TELKOMNIKA.v17i6.12701.
[9] M. A. Rasyidi and T. Bariyah, “Batik pattern recognition using convolutional neural network,” Bulletin of Electrical Engineering
and Informatics, vol. 9, no. 4, pp. 1430–1437, 2020, doi: 10.11591/eei.v9i4.2385.
[10] M. A. Rasyidi, R. Handayani, and F. Aziz, “Identification of batik making method from images using convolutional neural
network with limited amount of data,” Bulletin of Electrical Engineering and Informatics, vol. 10, no. 3, pp. 1300–1307, 2021,
doi: 10.11591/eei.v10i3.3035.
[11] A. H. Rangkuti, A. Harjoko, and A. Putra, “A novel reliable approach for image batik classification that invariant with scale and
rotation using MU2ECS-LBP algorithm,” Procedia Computer Science, vol. 179, pp. 863–870, 2021, doi:
10.1016/j.procs.2021.01.075.
[12] A. H. Rangkuti, A. Harjoko, and A. E. Putra, “Improvement of accuracy in batik i mage classification due to scale and rotation
changes using M2ECS-LBP algorithm,” Journal of Theoretical and Applied Information Technology, vol. 97, no. 14, pp. 3859–
3870, 2019.
[13] M. Alkaff, H. Khatimi, N. Lathifah, and Y. Sari, “Sasirangan motifs classification using scale- invariant feature transform (SIFT)
and support vector machine (SVM),” MATEC Web of Conferences, vol. 280, p. 5023, 2019, doi:
10.1051/matecconf/201928005023.
[14] I. Nurhaida, A. Noviyanto, R. Manurung, and A. M. Arymurthy, “Automatic Indonesian’s Batik pattern recognition using SIFT
approach,” Procedia Computer Science, vol. 59, pp. 567–576, 2015, doi: 10.1016/j.procs.2015.07.547.
[15] A. C. Siregar and B. C. Octariadi, “Classification of sambas traditional fabric ‘Kain Lunggi’ using texture feature,” IJCCS
(Indonesian Journal of Computing and Cybernetics Systems), vol. 13, no. 4, p. 389, 2019, doi: 10.22146/ijccs.49782.
[16] I. Nurhaida, R. Manurung, and A. M. Arymurthy, “Performance comparison analysis features extraction methods for Batik
recognition,” 2012 International Conference on Advanced Computer Science and Information Systems, ICACSIS 2012 -
Proceedings, pp. 207–212, 2012.
[17] F. Albert, J. M. Gómis, J. Blasco, J. M. Valiente, and N. Aleixos, “A new method to analyse mosaics based on symmetry group
theory applied to Islamic geometric patterns,” Computer Vision and Image Understanding, vol. 130, pp. 54–70, 2015, doi:
10.1016/j.cviu.2014.09.002.
[18] O. Chaowalit and P. Kuntitan, “Using deep learning for the image recognition of motifs on the Center of Sukhothai Ceramics,”
Current Applied Science and Technology, vol. 22, no. 2, 2021, doi: 10.55003/cast.2022.02.22.002.
[19] Y. Ma and O. Arandjelović, “Classification of ancient roman coins by denomination using colour, a forgotten feature in automatic
ancient coin analysis,” Sci, vol. 2, no. 2, p. 37, 2020, doi: 10.3390/sci2020037.
[20] F. Öztürk and F. Özen, “A new license plate recognition system based on probabilistic neural networks,” Procedia Technology,
vol. 1, pp. 124–128, 2012, doi: 10.1016/j.protcy.2012.02.024.
[21] P. Meghana, S. Sagar Imambi, P. Sivateja, and K. Sairam, “Image recognition for automatic number plate surveillance,”
International Journal of Innovative Technology and Exploring Engineering, vol. 8, no. 4, pp. 9–12, 2019.
[22] X. Di and V. M. Patel, “Deep tattoo recognition,” IEEE Computer Society Conference on Computer Vision and Pattern
Recognition Workshops, pp. 119–126, 2016, doi: 10.1109/CVPRW.2016.22.
[23] Z. H. Sun, J. Baumes, P. Tunison, M. Turek, and A. Hoogs, “Tattoo detection and localization using region-based deep learning,”
Proceedings - International Conference on Pattern Recognition, vol. 0, pp. 3055–3060, 2016, doi: 10.1109/ICPR.2016.7900103.
[24] R. T. Silva and H. S. Lopes, “A transfer learning approach for the tattoo detection problem,” pp. 1–8, 2021, doi:
10.21528/cbic2021-34.
[25] K. Mikolajczyk and C. Schmid, “A performance evaluation of local descriptors,” IEEE Transactions on Pattern Analysis and
Machine Intelligence, vol. 27, no. 10, pp. 1615–1630, 2005, doi: 10.1109/TPAMI.2005.188.
[26] S. Leutenegger, M. Chli, and R. Y. Siegwart, “BRISK: binary robust invariant scalable keypoints,” Proceedings of the IEEE
International Conference on Computer Vision, pp. 2548–2555, 2011, doi: 10.1109/ICCV.2011.6126542.
Iban plaited mat motif classification with adaptive smoothing (Silvia Joseph)
850 ISSN: 2252-8938
[27] E. Mair, G. D. Hager, D. Burschka, M. Suppa, and G. Hirzinger, “Adaptive and generic corner detection based on the accelerated
segment test,” Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes
in Bioinformatics), vol. 6312 LNCS, no. PART 2, pp. 183–196, 2010, doi: 10.1007/978-3-642-15552-9_14.
[28] A. Vedaldi and B. Fulkerson, “Vlfeat - an open and portable library of computer vision algorithms,” MM’10 - Proceedings of the
ACM Multimedia 2010 International Conference, pp. 1469–1472, 2010, doi: 10.1145/1873951.1874249.
BIOGRAPHIES OF AUTHORS
Ts. Dr Hamimah Ujir is a Senior Lecturer at the Faculty of Computer Science and
Information Technology (FCSIT), Universiti Malaysia Sarawak (UNIMAS). She obtained her PhD
in the School of Electronic, Electrical and Computer Engineering, University of Birmingham,
United Kingdom in 2013. Hamimah’s research interests lay in the interdisciplinary field of
computer vision, with related interests being in computer graphics, image processing, and
mathematical methods. Her previous and current works include 3D physical simulation and 3D
static as well as dynamic facial expression. She can be contacted at email: uhamimah@unimas.my.